Our compliance baseline fits in three sentences. Vee Center Sp. z o.o., the company that runs our services, holds ISO/IEC 27001:2022 certification. Personal data in our services is processed within the EU/EEA, with the client as controller and us as processor. And on the AI Act, we build transparency into every conversation and help each client assess its own configuration, rather than claiming a blanket conclusion on its behalf.
Procurement and compliance teams in banks, insurers and healthcare providers ask us about these things in every tender, and they are right to. This note sets out what each element means, what it does not mean, and why we avoid blanket compliance claims.
What is certified, and for whom#
The certificate belongs to Vee Center Sp. z o.o., the group company that holds corporate sales, infrastructure and research and development, and that delivers the service. Its information security management system was first certified to ISO/IEC 27001 on 4 November 2021, audited by SGS. The current certificate, to the 2022 edition of the standard, is valid from 4 November 2024 to 3 November 2027, subject to positive surveillance audits. Its scope is automatic customer service in the telephone channel.
Two precise points matter. First, the certificate applies to Vee Center Sp. z o.o., not to Vee S.A. or the group as a whole, so we do not write "Vee is ISO certified" without naming the entity. Second, a certificate is evidence of a managed system, not a promise that nothing can go wrong. What it shows is that risks are assessed, controls are defined and tested, and the system is audited internally and externally.
Behind the certificate sit the things clients' security teams usually ask to see: an information security policy, risk management, network security, incident management, data protection, business continuity and disaster recovery procedures that are tested, hardening and patching, internal audits, a register of authorisations, and technical and organisational measures mapped to the standard and to Article 32 of the GDPR.
Where data is processed, and whose it is#
Personal data in our services is processed within the EU/EEA. The contractual model is that the client is the data controller and Vee is the processor, set out in a data-processing agreement. In a bank's collections process, for example, we act as a sub-processor calling as part of the bank's own operation, and the bank keeps the legal basis for contact and decides which accounts are assigned to us.
- No mixing across clients. We do not combine personal data from different clients or projects. Customer data is a line we do not cross.
- Access on a need-to-know basis. Recordings are available to authorised people, such as our QA team listening to samples, within the terms of the processing agreement.
- Retention by agreement. Recordings and records are kept for as long as the project's rules allow, and deleted when they require it.
- Your access controls. Where a bank requires single sign-on and two-factor authentication for access to our panels, we integrate with its system.
- Audits. Client audits are designed remote-first, with on-site audits possible where remote evidence is not enough. Banks do audit us, across security policies, GDPR, the AI Act and DORA.
What the AI Act asks now#
Since 2 August 2026, Article 50 of the AI Act has required that people are informed when they are interacting with an AI system. The European Commission published guidelines on these transparency obligations on 20 July 2026. Our AI always introduces itself as AI, and that disclosure is designed into the opening of each conversation rather than added as a disclaimer.
This is not a new habit forced on us. In 2016, speaking to Puls Biznesu at the time of our stock-market debut, we said we would not hide from customers that they were talking to a machine. Our own code of conduct says the same: our AI never conceals its identity, however natural it sounds.
Disclosure done well does not cost the conversation. The way the AI introduces itself, the organisation it names and the moment it moves to the purpose of the call are designed and tested like any other part of the dialogue, so that customers know who they are talking to and still have a reason to stay on the line.
Beyond transparency, the Act does not classify a customer-service process by date or by label. The obligations depend on what the system actually does: whether it generates content on its own, whether it makes or recommends decisions with material consequences for a person, whether its behaviour changes after deployment without controlled change, and whether what happened can be reconstructed. Further obligations for high-risk uses apply on the timeline the Commission publishes. Two processes from the same supplier can therefore sit in very different places.
Compliance has been part of the process for a long time#
None of this was bolted on for the AI Act. In May 2018, as the GDPR took effect, one of our early outbound processes was built specifically to collect customers' GDPR consents by phone: a free conversation, then a clear yes or no confirmation. In 2021 Vee Center was first certified to ISO/IEC 27001. In our 2025 management report we described the AI Act as raising the value of auditable, reconstructable solutions, and that is how we have designed for regulated clients since.
In practice, the paperwork arrives early in every project, before any specification is written. A new deployment starts with the contract and the data-processing agreement, including sub-processing where it applies, and the client's data protection impact assessment questionnaires. Calling windows, attempt limits, retry intervals and recording-consent handling are then configured per market and enforced by the system, so that the rules agreed on paper are the rules that run. When we enter a new country, those settings are reviewed against local requirements rather than copied from the last market.
DORA and GDPR: whose obligations are whose#
For our financial-sector clients, DORA obligations, including incident reporting to supervisors, rest with the regulated entity. Our role is to provide the information and evidence within our control: incident classification by impact, notification under the contract, continuity procedures, and cooperation in audits. Under the GDPR, the client as controller decides the purposes and legal basis of processing; we process on its instructions and support it with the records it needs. We are careful not to blur these lines, because a supplier that claims to carry its client's regulatory obligations is promising something it cannot deliver.
How we help clients assess their configuration#
What we can offer is a structured technical and operational analysis of the specific process, so that the client's legal, data-protection and risk teams can reach their own conclusions with good information. It is not a law-firm opinion. It is an architecture-first review.
| Workstream | Question answered | Typical output |
|---|---|---|
| Architecture map | Which components are deterministic, statistical, generative or decision-supporting? | System boundary diagram and component inventory |
| Customer-impact map | What does the system say, decide, capture or influence? Does it affect a material outcome? | Process map and register of decision boundaries |
| Control and evidence design | How are content, versions, testing, escalation, logs and human intervention controlled? | Evidence and governance checklist |
| Risk-reduction options | Which design choices reduce unnecessary compliance cost while keeping the process effective? | Recommendations with alternatives |
| Client handover | What needs confirmation from legal, the DPO, security, model risk or the business owner? | Decision log and client action list |
An operational and architectural analysis, not legal advice.
In regulated configurations the architecture itself does much of the work. Customer-facing utterances come from a closed, approved and versioned catalogue; no new customer-facing language is generated on its own; no materially consequential decision is made autonomously; nothing changes in production without a versioned release; and every call can be reconstructed from the recording, logs, statuses and script version. Statistical components, such as intent classification, are separated from content and from consequential decisions, and assessed component by component.
Why we never claim blanket compliance#
You will not find a claim of complete compliance with the AI Act, the GDPR and DORA in anything we publish. It is too broad to be true of any supplier. Compliance is a property of a specific process, in a specific configuration, for a specific purpose, operated by a specific controller, and much of it rests with the client. A supplier can design for it, evidence it and support it. It cannot declare it on the client's behalf.
That sentence is less catchy, but it is accurate, and in our experience compliance teams trust it more. They have seen the slogans. What they want is a supplier who can show the certificate with the right entity on it, name where the data sits, explain the architecture component by component, and sit on the same side of the table during the assessment.
What to ask any supplier#
- Which legal entity holds the ISO/IEC 27001 certificate, what is its scope, and until when is it valid?
- Where exactly is personal data processed, and under which processing agreement and role allocation?
- How is AI disclosure built into the conversation, not just the terms and conditions?
- Which components generate content or make consequential decisions, and how is that evidenced?
- What evidence can you provide for our own DORA and GDPR obligations, and how are audits run?
- If you claim complete compliance, with what exactly, for which configuration, and who assessed it?
Good suppliers answer these questions with documents, not adjectives: a certificate naming the entity and its validity, a processing agreement naming the location and roles, an architecture description naming each component, and a sample call reconstructed in front of you. That is the standard we try to meet in every tender, and the one we would encourage you to hold everyone to, including us.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS